Fewer than half of FDA-authorized artificial intelligence devices for Alzheimer's disease disclosed basic demographic details like age, sex, or race in their training or validation data, according to a recent research letter.

Investigators evaluated the demographic transparency of artificial intelligence (AI)– and machine learning (ML)–enabled devices authorized by the FDA for Alzheimer's disease and related dementias (ADRD). They aimed to assess the availability and representativeness of training and validation data sets supporting device authorization.

Twenty-four FDA-authorized AI/ML-based devices for ADRD from January 2015 through December 2024 were identified. For devices with multiple authorizations, the first authorization was selected, except for one device where the second authorization served as a comparator. Data sources included FDA approval summaries, PubMed, manufacturer websites, and peer-reviewed publications.

Training data were disclosed in FDA summaries for 10 devices (41.7%) and in peer-reviewed publications for 5 devices (20.8%). Among FDA summaries, three reported disease status, two reported age, four reported sex, and none reported race or ethnicity. In contrast, peer-reviewed articles disclosed disease status for five devices, age for four, sex for five, and race or ethnicity for only one. Validation data were even less complete: only 2 FDA summaries (8.3%) and 10 peer-reviewed articles (41.7%) included any validation demographics. For 23 of the 24 devices, demographic reporting across key domains was incomplete, and no justifications were provided.

Of the 24 devices, 22 (91.7%) were cleared via the 510(k) pathway and 2 (8.3%) via De Novo classification. Most devices (n = 20; 83.3%) were intended for volumetric quantification of brain structures using magnetic resonance imaging, with smaller proportions targeting functional cognitive testing (n = 3; 12.5%) or amyloid quantification (n = 2; 8.3%). Radiology panels reviewed 21 devices (87.5%), while neurology panels reviewed 3 (12.5%).

These findings show limited transparency in the data supporting FDA authorization of AI/ML-based ADRD devices. The researchers also highlighted that insufficient demographic reporting may hinder equitable clinical application and risk perpetuating disparities in ADRD care.

Full disclosures can be found in the published research letter.

Source: JAMA